SSSaket Sumaninagenticmusings.hashnode.dev·Sep 30 · 19 min readDatabricks Preview & Beta Features Worth Watching in 2026 A practical, ranked guide to the most relevant new capabilities — with official documentation and the use cases they fit Reviewed against Databricks documentation and September 2026 platform release n00
SSSaket Sumaninagenticmusings.hashnode.dev·Sep 30 · 34 min readJev Engineering for Coding-Agent HarnessesTL;DR Jev is TypeSafe AI's "System One" model — not a chat model, not a replacement for the LLM in a coding agent. It returns bounded, typed judgments (Choice, Score, Noul) that software can consume 00
SSSaket Sumaninagenticmusings.hashnode.dev·Sep 30 · 33 min readAgent Memory for Agentic SystemsBuilding anomaly detection and churn prediction agents that learn, with mem0, Hindsight, and cognee TL;DR Stateless agents re-read the entire conversation every turn, so cost grows quadratically w00
SSSaket Sumaninagenticmusings.hashnode.dev·Sep 30 · 50 min readAI Agent Context Degradation in 2026: What's Changed Since Chroma's ReportA candidate mechanism for long-context degradation, the fixes proposed in 2026, and how GPT-6 Astra and Claude 5.5 respond to it TL;DR Chroma's 2025 report showed that LLM accuracy degrades as input 01N
SSSaket Sumaninagenticmusings.hashnode.dev·Sep 12 · 43 min readAgentic Benchmarks Explained: Measuring AI That Actually WorksFrom smart answers to verified outcomes For years, AI evaluation asked a narrow question: Can the model produce the right answer? That question still matters. Models are tested on mathematics, codin11G
